-
Notifications
You must be signed in to change notification settings - Fork 0
Reference MCP Summarize Artifact
Produce a grounded narrative summary of a single artifact (tablet).
| Argument | Type | Required | Default | Description |
|---|---|---|---|---|
p_number |
string | yes | — | CDLI P-number, e.g. P227657
|
focus |
string | no | general |
general, research, or translation_status
|
focus options:
-
general— overview of the tablet: period, genre, state of documentation, what the text contains if known -
research— emphasizes bibliographic context: editions, scholars, citation history -
translation_status— emphasizes what is and is not translated, what is lemmatized, what remains unknown
The first call to summarize_artifact for a given P-number is slower (~3–5 seconds) — the API calls Claude internally to generate the summary, then caches it in the agent_outputs table. Subsequent calls return the cached result immediately.
The cache is invalidated when any annotation_run cited by the summary is superseded by newer data. This means if a lemmatization is corrected or new ATF data arrives, the next summary request will regenerate.
When pipeline_completeness ≤ 2 (the tablet has no ATF and no lemmatization), and similar tablets exist in the corpus (same period, provenience, genre), the response includes a best-guess section clearly framed as hypotheses:
- "Based on tablets with similar metadata from this period and provenience, this may be..."
- "The genre and period are consistent with administrative records of the type..."
Every claim in the best-guess section cites the tablets it draws from. Nothing is stated as fact when it is inference.
ToolResponse[CardPayload]:
{
"data": {
"title": "P227657 — KTT 188",
"sections": [
{
"heading": "Overview",
"body": "Administrative clay tablet from the Ur III period (ca. 2112–2004 BC), provenience Ur..."
},
{
"heading": "Text content",
"body": "The tablet records grain rations (ninda, beer) in a school exercise format..."
}
]
},
"summary": "Ur III administrative tablet; word list (school exercise), fully transliterated.",
"sources": [
{ "label": "CDLI catalog", "annotation_run_id": 1 },
{ "label": "CDLI ATF", "annotation_run_id": 3 }
],
"interaction_id": "43"
}Source: github.com/wittkensis/glintstone · Issues · Edit this wiki
Start here
Getting Started
Overview
Data Model
- Data Sources
- Data Quality
- Data Issues
- Import Pipeline Guide
- ML Integration
- Citation Pipeline Summary
Reference — Data Model
Reference — API
Reference — MCP
Opportunities
Personas
Project
Research